Shaxzodbek Qambaraliyev

Secure-by-default FastAPI->MCP bridge. Deny-by-default exposure, scope->tool auth that fails closed (even on STDIO), built-in rate limiting, PII redaction at the source, and a structured audit log — in-process, no gateway.

Vital signs for your MCP server: inspect capabilities, benchmark tool-call latency (p50/p95/p99), and assert health in CI. The ab/k6/pytest for MCP servers — non-interactive, scriptable, LLM-free.

A drop-in confidence gate for LLM agents: turn token logprobs into an accept / verify / ask-a-human routing decision, with honest per-model calibration.

Agent memory in a single SQLite file — no vector DB, no server, no cloud. Keyword recall on stdlib alone; torch-free hybrid semantic search via model2vec + sqlite-vec. Python lib + MCP server + CLI.

Local-first LLM router: keep sensitive data and bulk text labor on your own models, call the cloud only for the hard part. Privacy-routing (fail-closed) + manager-worker delegation + allowlist guard, MCP-native.

Fast, deterministic, zero-API linter for AI agent prompt assets — SKILL.md, sub-agents, MCP tool descriptions, slash commands, and system prompts. Catch weak triggers, contradictions, token bloat, and broken frontmatter before they ship.

Turn a codebase, package, or doc into a valid Claude SKILL.md — offline, deterministic, and valid by construction. Optional Claude refinement. Zero runtime dependencies.

Lint Claude Agent SKILL.md files against Anthropic's exact spec — frontmatter, naming, description trigger-quality, and token budgets — in your editor, on the CLI, and in CI. Zero-config.